GEO Lab #03 · S3-CHATGPT-M1-R3 · Original response

This page presents one selected response from GEO Lab #03. It adds a smaller reading and citation entry point; it does not replace the full archive or collect a new answer.

The OpenAI arm uses the Responses API with web search and does not represent consumer ChatGPT Search. Grok records identify their actual consumer-interface mode and available model label.

Recorded conditions

Record ID
S3-CHATGPT-M1-R3
Platform
OpenAI API + web search
Question / condition
M1
Repeat
3
Model label
gpt-5.6-sol
Visible mode
Not independently recorded
Reasoning
medium
Submitted / first recorded time
2026-11-02T09:37:52.892373+00:00
Completeness
Recorded body complete
Search
Actual search activity observed; exact controls and evidence remain in the raw record.

Original prompt

I'm a solo SaaS founder with no GEO budget. How can I reliably measure whether ChatGPT and Grok recommend my product to potential buyers?

The agreed prompt is reproduced exactly. Any recorded difference in the actual submitted text or search UI is retained in the raw record and full appendix.

Original answer

You can measure this with **a repeatable prompt panel plus referral/conversion tracking**. Don’t treat one ChatGPT or Grok answer as a “ranking”—answers vary by personalization, location, search mode, and randomness. ChatGPT may use memory and approximate location; Grok may personalize from account activity and interests. ([help.openai.com](https://help.openai.com/en/articles/9237897-chatgpt-search))

## 1. Create a buyer-intent prompt panel

Build 10–20 prompts from questions actual prospects ask. Include several intent types:

| Intent | Example |
|---|---|
| Category discovery | “What are the best tools for **[job]**?” |
| Comparison | “Alternatives to **[competitor]** for **[persona]**?” |
| Requirements | “I need **[feature]**, **[integration]**, and a budget under **[$X]**.” |
| Pain-driven | “How can a **[persona]** solve **[problem]** without **[constraint]**?” |
| Switching | “Should I switch from **[competitor]** if I need **[differentiator]**?” |
| Shortlisting | “Recommend three **[category]** tools for **[specific use case]**.” |

Use language buyers would naturally use. **Do not include your product’s name** except in separate branded-accuracy tests.

Weight each prompt:

- **3:** high purchase intent
- **2:** active research
- **1:** broad discovery

## 2. Run a controlled weekly test

For both ChatGPT and Grok:

1. Start a new conversation.
2. Disable memory/personalization where possible.
3. Avoid custom instructions.
4. Use the same prompt wording.
5. Run each prompt **three times**, preferably on different days.
6. Record the date, product/model, whether web search occurred, answer, and cited sources.

Three repetitions matter because reasoning-model outputs aren’t deterministic; xAI’s own evaluation guidance says a single run can mislead and recommends reporting results across repeated runs. ([docs.x.ai](https://docs.x.ai/developers/model-capabilities/text/best-practices))

If 120 weekly queries is excessive, start with:

- 10 prompts
- 2 repetitions
- 2 assistants

That is **40 answers per week**.

## 3. Score the results objectively

Use a free Google Sheet with these columns:

```text
Date
Assistant
Prompt ID
Buyer segment
Intent weight
Web search used?
Product mentioned?
Explicitly recommended?
Recommendation position
Linked/cited?
Description accurate?
Relevant landing page cited?
Competitors recommended
Answer URL or screenshot
```

Score each answer:

- **0:** absent
- **1:** mentioned, but not recommended
- **2:** recommended as an option
- **3:** top-three recommendation
- **4:** primary recommendation
- **−1:** mentioned inaccurately or negatively

Track these headline metrics:

### Recommendation rate

```text
Runs explicitly recommending you ÷ total relevant runs
```

### Top-three share

```text
Runs placing you in the top three ÷ total relevant runs
```

### Citation rate

```text
Runs linking to your site ÷ total relevant runs
```

### Accuracy rate

```text
Accurate mentions ÷ all mentions
```

### Consistency rate

```text
Prompts where you appear in at least 2 of 3 runs ÷ total prompts
```

Your **consistency rate** is more useful than an isolated mention.

For an overall score, multiply each answer’s 0–4 score by its prompt’s intent weight. This prevents a vague educational query from counting as much as “Recommend a product I can buy today.”

## 4. Separate “model memory” from current web visibility

Run two versions of your most important prompts:

1. **Normal:** “Recommend tools for…”
2. **Search-explicit:** “Search the web and recommend current tools for…”

This tells you whether:

- The model already knows your product.
- Current web sources make it discoverable.
- Your product appears only when a specific page gets cited.

ChatGPT’s web-search answers can include citations, but OpenAI warns that search results and citations can be incomplete, outdated, or incorrect. ([help.openai.com](https://help.openai.com/en/articles/9237897-chatgpt-search))

Record which third-party pages are cited. Often the actionable insight is not “publish more blog posts,” but rather:

> “ChatGPT repeatedly relies on three comparison pages, and none includes us.”

## 5. Track actual buyer exposure

Prompt testing measures **recommendation availability**, not how many real buyers saw it. Add three free signals.

### ChatGPT referral traffic

In GA4, Plausible, PostHog, or server logs, create a segment for:

```text
utm_source = chatgpt.com
```

OpenAI says ChatGPT automatically adds `utm_source=chatgpt.com` to referral URLs from search results. ([help.openai.com](https://help.openai.com/en/articles/12627856-publishers-and-developers-faq%23%3A~%3Atext%3DChatGPT%20automatically%20includes%20the%20UTM%20parameter%20utm_source%3Dchatgpt.com%20in%20referral%20URLs%2C%20enabling%20clear%20tracking%20and%20analysis%20of%20inbound%20traffic%20from%20ChatGPT%20search%20results))

Track:

- Sessions
- Signup rate
- Trial-to-paid rate
- Landing pages
- Assisted conversions

### Grok referral traffic

Create analytics filters for referrers containing:

```text
grok.com
x.com
```

Treat this as a **lower-bound signal**, not a complete count: links may be copied, referrer data may be stripped, and not every recommendation generates a click.

### Self-reported attribution

Add this required or optional signup question:

> How did you first hear about us?

Options:

- Google/search engine
- ChatGPT
- Grok
- Another AI assistant
- Social media
- Recommendation
- Other

Also add an optional text field:

> What did you ask or search for?

This is often your best source of new monitoring prompts.

## 6. Use three levels of evidence

Report results in this order:

1. **Visibility:** Your controlled prompt panel recommends you.
2. **Engagement:** Analytics shows visitors from ChatGPT/Grok.
3. **Revenue:** Signups or customers attribute discovery to them.

Don’t claim that “ChatGPT sends buyers” based solely on your own prompt tests. A reliable statement looks more like:

> “Across 60 controlled high-intent runs, we were recommended 28% of the time, linked 12% of the time, and received five attributed trials.”

## Recommended no-budget cadence

**Every week, 30–45 minutes:**

- Run your ten highest-intent prompts twice.
- Save outputs or screenshots.
- Update recommendation, top-three, citation, and accuracy rates.
- Check AI referral traffic.
- Add new buyer wording from sales conversations.
- Investigate pages ChatGPT and Grok cited instead of you.

**Every month:**

- Remove unrealistic prompts.
- Add emerging competitors.
- Compare the latest four-week average with the previous four weeks.
- Review branded prompts for factual errors.

Also confirm that your site does not block `OAI-SearchBot`; OpenAI states this is necessary for eligibility in ChatGPT search, though inclusion and placement are not guaranteed. ([help.openai.com](https://help.openai.com/en/articles/9237897-chatgpt-search))

The practical goal isn’t “know every answer every buyer receives”—that’s impossible. It is to maintain a **stable, buyer-weighted sample that detects meaningful changes and connects them to traffic and revenue**.

Recorded sources

These are the source roles recorded in the public dataset. A retrieval or source-panel item is not automatically an answer-body citation. A URL supplied in the prompt and repeated by an answer does not prove content access. No destination has been repaired or independently validated here.

Recorded answer-body links (5)

Retrieval-only / source-panel inventory (27)

Evidence and qualifications

This is the existing redacted public record. It has not been rewritten for this page. Earlier privacy edits and collection qualifications continue to apply. Editorial text is English; a few original captures retain non-English UI badges or source-panel date labels. An answer’s empirical claims are platform output, not verified findings of this archive.

Body review

{
"reviewed_at": "2026-11-02T09:39:03.953039+00:00",
"reviewer": "assistant",
"complete_body_read": true,
"coherent_ending": true,
"annotation_and_explicit_url_coverage_checked": true,
"external_claims_verified": false,
"independent_human_double_coding": false
}

Existing public-text privacy policy

{
"body_changed": true,
"policy": "Only actual URL tracking, signed/session URL parameters and necessary identifying information are removed; original wording, including analytics parameter examples outside URLs, remains.",
"annotation_positions": "Original position indexes omitted because URL/privacy edits can change character locations."
}

Prompt SHA-256: 709fd6fc82fe3f87b623fb3dd8b2adecdd6655c3508115f8b40c573d2397776d
Answer SHA-256: 17130e5f2f1dd04a7380bc330301848bec90a18392d789897cab3901820e6064